FlexCast Services: Virtualize 3D professional graphics Design Guide

Virtualize 3D
professional
graphics

citrix.com
FlexCast Services: Virtualize 3D professional graphics Design Guide 2

Table of contents

About FlexCast Services design guides 3

Project overview 4
Objectives 4
Virtualize 3D professional graphics 5
Solution components 5
Classification of 3D professional graphics users 7
Use case 1 – designers and engineers 8
Use case 2 – power users and operators 8
Solution architecture 9
User layer 10
Access layer 11
Resource layer 11
Control layer 12
Hardware layer 14
Validation 16
Summary 17
Appendix 17
References 20

p
Click on the section
names above to navigate
to that portion of the
book and the arrow icon
to return to the table of
citrix.com contents from any page.
FlexCast services: Virtualize 3D professional graphics Design Guide

About the FlexCast

Services design guide
Citrix FlexCast Services design guides
provide an overview of a validated
architecture based on many common
scenarios. Each design guide relies on
Citrix Consulting best practices and in-
depth validation by the Citrix Solutions
Lab to provide prescriptive design
guidance on the overall solution.
Each FlexCast Services design guide
incorporates generally available
products and employs a standardized
architecture, allowing multiple
design guides to be combined into
a larger, all-encompassing solution.
The design guide for virtual 3D
professional graphics spends more
time talking about the use-case and
different technologies for enabling the
graphics delivery. More information
on the related Citrix components are
available in one of the other guides.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 4

Project Overview
In an increasingly global economy, companies are looking to improve time to
market by securely collaborating and managing design lifecycles with offshore,
mobile and remote employees while maintaining secure control over intellectual
property (IP). Organizations are seeing client virtualization as an enabling
technology to accomplish these dual goals.

Citrix has introduced game-changing technology over the past half-decade, to

make virtualization of professional 3D graphics applications easy to deliver and
meet the performance expectations of designers and engineers. This is taken to a
whole new level with exclusive support for NVIDIA GRID virtual GPU’s in Citrix
XenServer. The benefits of moving graphics processing from under the desk to a
central datacenter are now well understood, both by large and medium
enterprises. Significantly, the technology from Citrix is able to address the variation
in cost and complexity requirements for different tiers of users within the
organization. That is a powerful stimulus in accelerating adoption of virtual 3D
graphics at scale.

The audience for this design guide is anyone already familiar with the physical
infrastructure for 3D graphics. It provides a starting point to understand the
technologies and scope of the project to transform that infrastructure using
Citrix virtualization.

Objectives
The objective of this FlexCast Services Design Guide is to construct and
demonstrate a way of globally delivering 3D professional graphics apps and 3D
data to enable real-time collaboration of design data, while securing IP.

The hypothetical organization in this example is called WorldWide Corporation

(WWCO), a large manufacturing firm with globally distributed design and
manufacturing centers that is currently supporting 3D professional graphics
applications globally through nightly time consuming file transfers of data to
multiple data centers which is an asynchronous method and slows real time
collaboration processes. Accessing 3D professional graphics applications hosted
in the data center using existing solutions is slow compared to running the
applications locally on workstations and inherently insecure. To address these
challenges, WWCO has decided to implement Citrix XenDesktop virtualized client
delivery platforms to resolve access performance, improve data security and
accelerate real-time global collaboration.

WWCO business objectives

• The desire to leverage a global talent base – Organizations recognize that to be

competitive, they need to leverage technical talent wherever it’s located. Reasons
include cost control and the ability to provide support close to the end customer.

• Corporate requirements to safeguard product design IP – The need to

share information with contractors, business partners and employees of
outsourcing services providers is driving organizations to improve protection of
their IP.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 5

• Workers’ need to view or present design models on mobile devices – The

ability to pull up design documents and sophisticated models on the shop floor or
at a customer location is becoming essential.

• Economic demands for cost control and faster time to market – Follow-
the-sun (24 x 7) development cycles and dispersed development teams require
real-time, remote collaboration on design data.

WWO Technical objectives:

• Single solution must accommodate user requirements of designers, engineers

and editors or viewers of professional 3D graphics.

• Build a solution which can scale from few hundred users to thousands of users.

• The solution must be validated and ready to be deployed within weeks

• Virtualize where possible to reduce costs and complexity.

• Implement n+1 highly available solution for business continuity

Virtualize 3D professional graphics

The design of this FlexCast Services is based on best practices from Citrix
consulting services and product development teams. The following assumptions
were made in determining these parameters:

• All users will access Windows-based 3D professional graphics applications via a

single datacenter, which will host all physical and virtual servers

• Applications require the latest OpenGL versions with GPU hardware acceleration.

• N+1high availability is required for physical components.

• Remote access is required for accessing Windows applications from outside

the firewall.

• WWCO’s existing infrastructure for Microsoft Active Directory, DNS/DHCP, and

Microsoft SQL Server will be reused.

Remember, this is a simplification for the purpose of understanding. A number of

factors must be considered, because end user experience in graphics technology
depends on current and expected workflows, network conditions, type and size of
the image files, and nature of 3D apps among many other parameters. You are
encouraged to work with the local Citrix partners and consultants to determine the
right mix of technologies for your environment.

Solution components
For the hardware, this guide considers NVIDIA GRID compatible servers that
support up to two NVIDIA GRID K2 cards per chassis. Each NVIDIA GRID K2 card
contains two onboard GPU’s, with 4GB frame buffer available to each GPU. GRID
boards feature the NVIDIA Kepler architecture that, for the first time, allows
hardware virtualization of the GPU. This means multiple users (virtual machines, in

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 6

this case) can share a single GPU, improving user density while providing true PC
performance and compatibility. The scale of sharing depends on use-case
requirements, which maps to different vGPU Types as per Table 1. vGPU are
analogous to physical GPU’s, having a dedicated and fixed amount of GPU
frame-buffer (reserved via the physical frame-buffer) and one or more virtual
display outputs or “heads”.

GRID Physical NVIDIA Suggested vGPUs vGPUs GPU

card GPUs per vGPU use-case per per memory
board type pGPU board per VM
GRID K2 2 Pass- Designer 1 2 4 GB
through
GRID Designer 2 4 2 GB
K260Q

GRID Power user 4 8 1 GB

K240Q
GRID Knowledge 8 16 256 MB
K200 worker
GRID K1 4 Pass- Power user 1 4 4 GB
through
GRID Power user 4 16 1 GB
K140Q
GRID K100 Knowledge 8 32 256 MB
worker
Table 1: vGPU types differ in amount of frame-buffer, virtual display heads, max resolution, and number of
simultaneous instances.

In addition to the NVIDIA GRID cards discussed here, Citrix supports other graphic
cards from NVIDIA and AMD. Please see this knowledge base article for details

The Citrix software components that make up the solution are as follows:

• Citrix XenDesktop 7.1

• Citrix XenServer 6.2

• NetScaler Gateway 10.1

• Citrix StoreFront Services 2.1

• Citrix Licensing Server 11.11

• Citrix Receiver 3.4 or higher

These Citrix components communicate with each other to deliver a secure

connection to virtualize desktops for 3D applications from devices located inside
and outside the corporate network. For an in-depth technical explanation of
component communication, please see Appendix 1.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 7

Classification of 3D professional graphics users

Designing a good solution requires analysis of the current and expected usage,
and classification of users according to their demands on the GPU. The following
list and figure provides a simplified recommendation to segment the 3D
professional graphics users at WWCO based on GPU requirements, and define the
right solution for each group.

1. Designers and engineers: The most demanding user group. They create and
manipulate large, complex, 3D models and require a dedicated GPU for
graphics acceleration.

2. Operators and contractors (Power users): Users are classified in this segment
when they need to view or edit graphics intensive 3D files, or access complex
graphics workflows onsite, say on the factory floor or at a construction site;
hardware GPU acceleration is recommended.

3. Knowledge and task workers: The segment of users in the organization that are
not engaged in professional graphics design. Hardware accelerated graphics
may or may not be required to deliver business graphics, such as the Windows
8 style apps, PowerPoint transitions in Office 2013, or perform light 2D and 3D
work. The cost and design considerations for business graphics are not in
scope of this guide (faded area in diagram)

This solution at WWCO consists of the following types of users:

Use Compute User group Number Example workflow

case requirement of users
1.A Ultra high-end Designers 12 Frequently create, edit, and render
complex and large 3D models
1.B High-end Engineers 24 Frequently edit and share these
large 3D models
2.A Mid-range, Operators 48 Perform complex workflows with
consistent multiple instances of graphics
applications, on factory shop floor.
2.A Mid-range, Contractors 116 Need to pull up detailed 3D
bursty models in graphics applications to
perform related engineering tasks

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 8

Workload analysis is necessary to determine the Scalability and Sizing during the
PoC testing. The following guidelines explain the concept making simplistic
assumptions about users with homogenous and constant workload within each
category. Two NVIDIA GRID K2 cards are used per server in each case, as per
appropriate vGPU types from Table 1. The proposed vGPU assignment and sizing
is discussed after Table 2

Use User group Number GRID K2 vGPU VM’s supported Total

case of users type per server (2 servers
slots)
1.A Designers 12 Pass-through 4 3
1.B Engineers 24 K260Q 8 3
2.A Operators 48 K240Q 16 3
2.B Contractors 116 Pass-through 4 (40 users) 3
(Windows server)
Redundancy N+1 One card failure 2

Table 2: Sizing recommendation based on the assumptions

Use case 1 – Designers and engineers

For ultra-high-end 3D compute requirements, such as the designers in use-
case 1.A, a dedicated desktop environment is made available to each user, while
the underlying hardware resources are shared using Citrix XenDesktop. With
XenServer GPU pass-through users share a single server but each user has 1:1
GPU assigned to them.

Shared GPU for desktops with high-end vGPU types such as K260Q may
be suitable for engineers (use-case 1.B) who have high end 3D compute
requirements, and perform graphics intensive operations on 3D models. 2:1 GPU
assignment doubles the user density per server.

An additional server with two GRID K2 cards is required to handle failure of any
one card on the primary hosts. Two servers and four GRID K2 cards are required
for full server redundancy.

Total servers for use-case 1, with card level redundancy = 3 + 3 + 1 = 7 servers

Use case 2 – Power users and operators

Operators (use-case 2.A) that need to pull up complex graphic models to do
their work, maybe at a construction site or a factory floor, still expect the images
to load fast, respond quickly, and maintain high-fidelity resolution. High-density
vGPU types, such as K240Q, are suitable for users that spend a large part of
their workday viewing and editing 3D files. Shared GPU for desktops allows
discrete GPU resources to be directly mapped to virtual machines with N:1 GPU
assignment for mid-range 3D compute performance.

Shared GPU for applications is the most cost-effective, high-performing solution

in the industry for high-density. The larger proportion of users at many design
organizations are not driving the compute resources all the time, but whenever

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 9

they do they expect the system to flawlessly bring up their 3D models. Their 3D
compute requirements are mid-range, but infrequent and spread through the
day. For such users, that we have named contractors (use-case 2.B) in this
example, deliver graphics apps from Windows Server sharing single GPU1 among
multiple user-sessions. N:1 GPU sharing cost effectively supports users that view
and edit 3D data and can adequately be supported by sharing GPU resources.
Customers have reported running 20, 30, and even 100 users in this way. Our
design conservatively assumes 10 users per GPU, which roughly represents viewer
workload on Autodesk AutoCAD. With a K2 card having two on-board GPUs, you
can scale up to 40 users per server.

Similar to use-case 1, failure of any card can be handled with an additional standby
server containing pair of GRID K2 cards. Configure each card to handle the VMs
for one of the above use-cases, respectively.

Total servers for use-case 2, with card level redundancy = 3 + 3 + 1 = 7 servers

Hardware specification and sizing for these use-cases is discussed in the

Resource Layer section. The infrastructure machines add very little incremental
load for this scenario. In this PoC, control virtual machines will be hosted on the
same servers as Resources.

Solution architecture
In the following sections, we look at the different parts of the Citrix infrastructure
to enable the 3D graphics solution. Figure 2, based on the overall business
and technical objectives for the project as well as the assumptions, provides a
graphical overview of the solution architecture.

Figure 2: Conceptual Diagram

While this guide considers 200 users, the core architecture design does not really
change. The redundancy and scalability features will support thousands of users.
The limiting factor is how many XenServer resource pools will be needed. For 200
users, a single cluster or resource pool is sufficient.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 10

The overall solution for WWCO is based on a standardized five-layer model,

providing a framework for the technical architecture. We don’t go into the details
of all Citrix components and architecture here, to keep the discussion limited to
the graphics technologies. The information here is sufficient to bring up a working
environment to deliver 3D graphics. For detailed recommendations and best-
practices for deploying the Citrix infrastructure, please see the FlexCast Services
guides referenced in Appendix

At a high level, the 5-layer model comprises:

1. User layer. Defines the unique user groups and overall endpoint requirements.
We explored this in the previous section.

2. Access layer. Defines how user groups will gain access to their resources.
Focuses on secure access policies and desktop/application stores.

3. Resource layer. Defines the virtual resources, which could be desktops or

applications, assigned to each user group. In this context, it means the
virtualized, GPU-accelerated graphics apps.

4. Control layer. Defines the underlying infrastructure required to support the users
in accessing their resources.

5. Hardware layer – Defines the physical implementation of the overall solution with
a focus on physical servers, graphics cards, storage and networking.

Figure 2: Virtual desktop model

User layer
The user layer focuses on the logistics of the user groups, which includes client
software, recommended endpoints and office locations. This information helps
define how users will gain access to their resources, which could be desktops,
applications or documents.

• Citrix Receiver client. This client software, which runs on virtually any device and
operating platform, including Windows, Mac, Linux, iOS and Android, must be
downloaded onto user endpoints to access graphics applications, which are hosted
in the datacenter. Citrix Receiver provides the client-side functionality to secure,
optimize and transport the necessary information to/from the endpoint/host over
Citrix HDX, a set of technologies built into a networking protocol that provides a
high-definition user experience regardless of device, network or location.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 11

• Endpoints. The physical devices could be smartphones, tablets, laptops,

desktops, thin clients, etc. Users download and install the Citrix Receiver client
from their device’s app store or directly from Citrix.com. For the graphics use-
case, choice of end-point depends on the requirements. For example, CAD
designer is not expected to use a tablet for designing but they will probably want
to use a tablet for reviewing purposes.

• Location. The system accounts for users that work from remote locations, over
unsecure network connections, requiring all authentication and session traffic to
be secured. Please review network latency and user-experience expectations
when working remotely or from a mobile device.

Access layer
The access layer defines the policies used to properly authenticate users to the
environment, secure communication between the user layer and resource layer
and deliver the applications to the endpoints.

Note: In an isolated proof of concept limited to the LAN of a lab environment, the
virtualized graphics delivery will work even without the access layer components.
In that case, you must ensure security compliance elsewhere in the network.

The following displays access layer design decisions based on WWCO requirements.

Users connecting from… Remote, untrusted network

Authentication point NetScaler Gateway
Authentication policy Multi-factor authentication
(username, password and token)

Allowing users to access the environment from a remote location without

authenticating would pose security risks to WWCO. When users access the
environment, the external URL will direct requests to Citrix NetScaler Gateway,
which is deployed within the DMZ portion of the network. NetScaler Gateway
will accept user multi-factor authentication credentials and pass them to the
appropriate internal resources (Active Directory domain controllers and token
authentication software such as RADIUS).

Resource layer
This layer manages the image, optimizations, and the delivery mechanism. This is
the most technically complex layer in the solution deployment. Virtual desktops,
hosted applications, or both, are delivered from this layer using XenDesktop and
XenApp software.

Citrix gives you two ways of delivering resources to your end-users:

a) Using XenApp or XenDesktop 7.1 apps, only the Windows apps are presented
from a Windows Server platform, occupying a smaller footprint on the client; or

b) Using XenDesktop 7.1, replicate the complete physical desktop including their
apps and “personalization” in a virtual environment based on Windows 7 or
Windows 8.1

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 12

Based on the requirements captured in the Solution Design section, the following
resource layer design decisions go into creating the Virtual Machine base image.
We need to create two different master images, one for Windows Server OS and
the other with Windows Desktop OS:

Criteria Decision for apps Decision for desktops

Operating system Windows Server 2008 R2 Windows 7 SP 1
Delivery Machine creation services Machine creation services
CPU 8 vCPU 4 vCPU
Memory 32 GB RAM 8 GB RAM
Disk 60 GB 60 GB
Application(s) 2
Autodesk AutoCad, Autodesk AutoCad,
SolidWorks, PTC Creo, SolidWorks, PTC Creo,
Siemens SolidEdge, etc. Siemens SolidEdge, etc.
Graphics acceleration GPU pass-through on NVIDIA GRID (vGPU) on
XenServer 6.2 XenServer 6.2
User group Power users (Contractors, Designers, engineers, power
operators) users
Number of VMs 12 84

Control layer
The control layer of the solution defines the virtual servers used to properly
deliver the prescribed environment detailed in the user, access, and resource
layers of the solution, including required services, virtual server specifications
and redundancy options.

Control layer components include access controllers, delivery controllers and

infrastructure controllers.
Access controllers3
The access controllers are responsible for providing users with connectivity to their
resources, as defined within the access layer. In order to support the access layer
design, the following components are required:

Parameter NetScaler Gateway StoreFront

Instances 2 virtual servers 2 virtual servers
CPU 2 vCPU 2 vCPU
Memory 2 GB RAM 4 GB RAM
Disk 3.2 GB 60 GB
Citrix product version NetScaler VPX for XenServer StoreFront 2.1
10.1
Microsoft product Not applicable Windows Server 2012 R2
version
Network ports 443 443
Redundancy High-availability pair Microsoft Network Load
Balancing (MAC spoofing)

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 13

The redundant pair of NetScaler Gateway virtual servers is responsible for

providing secure, remote access while the redundant pair of StoreFront virtual
servers is responsible for the portal where users can see and pick the apps or
desktops they want.
Delivery controllers
The delivery controllers manage and maintain the virtualized resources for
the environment. In order to support the resource layer design, the following
components are required:

Parameter XenDesktop delivery controller

Instances 2 virtual servers
CPU 2 vCPU
Memory 4 GB RAM
Disk 60 GB
Citrix product version XenDesktop 7.1
Microsoft product version Windows Server 2012 R2
Network ports 80, 443
Redundancy Load balanced via StoreFront

A single delivery controller can easily support far more than the load of 200 users.
However, to provide N+1 fault tolerance, a second virtual server will provide
redundancy in case one virtual server fails.
Infrastructure controllers
In order to have a fully functioning virtual desktop environment, a set of standard
infrastructure components are required.

Parameter SQL server Citrix license Active directory4

server
Instances 2 virtual servers 1 virtual servers 2 virtual servers
CPU 2 vCPU 2 vCPU 2 vCPU
Memory 4 GB RAM 4 GB RAM 4 GB RAM
Disk 60 GB 60 GB 60 GB
Version(s) Not applicable Citrix License Server Not applicable
11.12
Microsoft Windows Server 2012 R2 Windows Server Windows Server
product version SQL Server 2012 2012 R2 2012 R2
Network ports 1433 27000, 7279, 8082 Default
Redundancy SQL Server AlwaysOn None due to 30 day Primary and
grace period backup domain
controller

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 14

To provide fault tolerance, the following options were used:

• The XenDesktop database was deployed on an HA pair of Microsoft SQL

Server 2012 servers utilizing the AlwaysOn availability group with primary and
secondary instances spread across two virtual servers.

• Once active, a XenDesktop environment can continue to function for 30 days

without connectivity to the Citrix License Server. Due to the integrated grace
period, no additional redundancy is required.

Hardware layer
The hardware layer is the physical implementation of the solution. It includes server,
networking and storage configurations needed to successfully deploy the solution.
Server
Following is the physical server implementation for the WWCO solution. The
same hardware was leveraged for both control and resource layer to benefit from
economies of scale:

Component Description Quantity Total

Server model Cisco C240 M3 7+7 14 servers
Processor(s) Intel Xeon E5-2690 @2.9GHz 2 16 cores/ Server
Memory 16GB DDR3-1333 16 256 GB/ Server
Disk(s) 600GB SAS @ 15,000RPM 12 7.2 TB
Hypervisor Citrix XenServer 6.25
14 14
(with Service Pack 1 for vGPU)

To provide fault tolerance within the solution, the virtual servers were distributed
so redundant components were not hosted from the same physical server.
The resource load on the physical hardware for the access and control layer
components is minimal, which is why they are hosted on the standby resource
layer servers to optimize the return on investment (RoI). The virtual server allocation
is depicted in Figure 3.

Server 1 and Server 2 host the access and control layer components, in addition
to the VM’s connected to GPU. “RDS Host” VMs contain Windows Server OS,
while “Desktop VMs” contain Windows 7 OS.

Figure 3: Virtual machine server allocation.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 15

Storage
The storage architecture for the solution is based on the use of inexpensive local
storage. To ensure an acceptable user experience, the storage architecture must
have enough throughput capacity as well as fault tolerance to overcome the
potential failure of a single drive.

Parameter Resource Layer Hosts

Drive count 12
Drive speed 15,000 RPM
RAID RAID 5

Networking
Integrating the solution into the network requires proper configuration to have
the right components communicate with each other. This is especially important
for NetScaler Gateway, which resides in the DMZ. Large graphic image files can
consume bandwidth, so the network sizing must be done keeping use-case
requirements in mind. The network is configured based on each physical server’s
having four network ports:

NIC instance Function Speed VLAN ID

1 Management VLAN 1 Gbps or more 1
2 Virtual machine VLAN 1 Gbps or more 2
3 DMZ VLAN 1 Gbps or more 3
4 Disabled

The three VLANs are divided among the physical servers, NetScaler Gateway and
remaining virtual servers as shown in Figure 4.

Figure 4: Networking architecture

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 16

As depicted in the diagram, the VLAN is configured as follows:

• NetScaler Gateway is configured to use the DMZ VLAN. This VLAN does not
connect with any other internal networks, which helps keep the DMZ and internal
traffic separated.

• The management VLAN is only connected to the physical hosts and not the
virtual machines. This VLAN is for management calls to/from the physical
server’s hypervisor.

• The virtual machine VLAN, meant for all non-DMZ virtual machines, allows them
to connect to the internal datacenter network.

Validation
The solution was validated in the Citrix Solutions Lab using graphic benchmark
apps, running on the different GPU sharing models described above.

The charts below is taken when running Redway3D’s RedTurbine benchmark app
on multiple VM’s sharing the K240Q vGPU on a GRID K2. Notice that the frames-
per-second (FPS) chart in the baseline test (single user) looks very similar to the nth
user when GPU is shared close to 85%. This indicates performance is maintained
even when the GPU is being heavily used by all users at the same time.

Baseline 4 users simultaneously sharing GRID K2

Average performance is greater than 40 FPS With 4 users, average remains greater than 40 FPS;
momentary drop to 10 FPS in the beginning

Average load is about 30% of the total GPU With 4 users, average load is about 80-85%

Table 3: Performance of four Windows 7 VDI sessions sharing a GRID K2 GPU with XenDesktop 7.1

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 17

Summary
Many business benefits result from virtualizing 3D professional graphics apps,
going beyond the cost rationalization to leveraging worldwide talent pool while
securing intellectual property (IP), increasing productivity with flexible mobile device
access from anywhere, anytime, and gaining ability to respond quickly to line-of-
business requests.

The Citrix solution is mature and specially designed to support graphics intensive
apps and deliver an exceptional experience to designers and engineers and 3D
data viewers and editors working with these apps. Customers can leverage
existing Citrix investments, because no new XenDesktop or XenApp infrastructure
is required.

Many large product design, manufacturing, and engineering firms have

successfully deployed Citrix HDX 3D Pro solution for mission-critical projects and
are profiting from collaboration among their design engineers across the globe.
Citrix and its partners are ready to share experiences and best practices to help
you on this journey.

Appendix
Virtualize 3D professional graphics process overview
HDX 3D Pro integrates with your existing XenDesktop infrastructure and leverages
the same XenDesktop services such as provisioning services, profile management,
app streaming and Desktop Director. HDX 3D Pro supports both XenServer VMs
and physical host computers – including desktop, blade, and rack workstations.
Again, you can deliver graphical applications either as part of a complete virtual
desktop or as a VM hosted app.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 18

Step Process name Description

1 Virtual desktop agent
2 XenDesktop controller
3 Citrix Receiver/ Thin client
4 Connection brokering
5 Encoding
The architecture begins with a physical machine
or XenServer virtual machine (VM) hosting the
application where you install the Virtual Desktop
Agent for HDX 3D Pro.
From a catalog containing the computer hosting
the graphical app and a desktop group you
create, you assign the desktop or VM hosted app
to a user.
Users access the desktop or VM hosted app
through a Windows device or a XenDesktop-
compatible Linux thin client running the
appropriate Citrix Receiver.
When a user logs on to Citrix Receiver and
accesses the desktop or VM hosted app, the
controller authenticates the user and contacts the
Virtual Desktop Agent for HDX 3D Pro to broker a
connection to the computer hosting the graphical
application.
Next, the Virtual Desktop Agent for HDX 3D Pro
uses the appropriate hardware on the host to
compress views of the complete desktop or just
of the graphical application.

CPU-based Deep Compression codec is the

6 HDX 3D image delivery
default in latest version of HDX 3D Pro.
These views, and the user’s interactions with
them, are transmitted between the host computer
and the user device through a direct HDX
connection between Citrix Receiver and the
Virtual Desktop Agent for HDX 3D Pro.

Citrix shared GPU for desktops (HDX 3D Pro)

Extracted from NVIDA GRID vGPU user guide by Andy Currid

Citrix shared GPU for desktops (HDX 3D Pro) is true hardware-accelerated

graphics for every virtual machine (VM), equivalent to the graphics stack available
on physical workstations. Citrix HDX 3D Pro uses the native graphics card driver
installed directly in the guest OS. The complete NVIDIA stack (NVIDIA hardware,
NVIDIA guest OS drivers, and NVIDIA GRID manager) on XenServer ensures that
applications can leverage 100% features on the GPU card, including the latest
OpenGL 4.3 and DirectX 11 libraries.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 19

Shared GPU with Citrix XenServer (vGPU)

GRID vGPU’s high-level architecture is illustrated in the figure above. Under the
control of NVIDIA’s GRID Virtual GPU Manager running in XenServer dom0, GRID
physical GPUs are capable of supporting multiple virtual GPU devices (vGPUs) that
can be assigned directly to guest VMs.

Guest VMs use GRID virtual GPUs in the same manner as a physical GPU that
has been passed through by the hypervisor: an NVIDIA driver loaded in the guest
VM provides direct access to the GPU for performance-critical fast paths, and
a paravirtualized interface to the GRID Virtual GPU Manager is used for non-
performant management operations.

GRID vGPUs are analogous to conventional GPUs, having a fixed amount of

GPU framebuffer, and one or more virtual display outputs or “heads”. The vGPU’s
framebuffer is allocated out of the physical GPU’s framebuffer at the time the vGPU is
created, and the vGPU retains exclusive use of that framebuffer until it is destroyed.

All vGPUs resident on a physical GPU share access to the GPU’s engines including
the graphics (3D), video decode, and video encode engines.

citrix.com p
FlexCast Services: Virtualize 3D professional graphics Design Guide 20

References
• Reviewer’s guide for delivering 3D graphics (vGPU)
http://www.citrix.com/skb/articles/RDY12202

• Installation and configuration guide for HDX 3D Pro

http://support.citrix.com/proddocs/topic/xendesktop-als/hd-3d-install.html

• FlexCast Services design guides

https://www.citrix.com/solutions/desktop-virtualization/overview.html

• FlexCast Services guide for mobilizing Windows apps

http://deliver.citrix.com/WWWB0513XDDGUIDEMOBILEAPPSWP.html

• XenDesktop 7 handbook
http://support.citrix.com/article/CTX139331

• Citrix tested hardware for HDX 3D Pro

http://support.citrix.com/article/CTX131385

• Citrix tested hardware for XenServer GPU pass-through

http://hcl.xensource.com/GPUPass-throughDeviceList.aspx

• NVIDIA GRID certified servers

http://www.nvidia.com/buygrid

citrix.com p
 FlexCast Services: Virtualize 3D professional graphics Design Guide 21

• NVIDIA GRID vGPU user guide

http://www.nvidia.com/vgpu

• Citrix XenServer 6.2 with vGPU feature pack release notes

http://www.citrix.com/go/vgpu

• HDX 3D Pro technical FAQs

http://www.citrix.com/skb/articles/RDY2465

1. GPU Pass-through to a Windows Server VM. Multiple users can launch GPU-accelerated 3D application
sessions, using Citrix XenApp or Citrix XenDesktop 7 App Edition, all sharing the same physical GPU
2. Applications will be defined by the customer. This is just a random sampling of graphics apps.
3. If you choose not to implement the access layer security component, only the StoreFront servers are required.
4. To simplify the PoC, and never in production, the AD, DNS and DHCP services may be installed on the same VM
5. In tech preview at time of writing. Version number may change when general availability of vGPU is announced

Corporate Headquarters India Development Center Latin America Headquarters

Fort Lauderdale, FL, USA Bangalore, India Coral Gables, FL, USA

Silicon Valley Headquarters Online Division Headquarters UK Development Center

Santa Clara, CA, USA Santa Barbara, CA, USA Chalfont, United Kingdom

EMEA Headquarters Pacific Headquarters

Schaffhausen, Switzerland Hong Kong, China

About Citrix
Citrix Systems, Inc. (NASDAQ:CTXS) is a leading provider of virtual computing solutions that help companies deliver IT as an on-demand service.
Founded in 1989, Citrix combines virtualization, networking and cloud computing technologies into a full portfolio of products that enable virtual
workstyles for users and virtual datacenters for IT. More than 230,000 organizations worldwide rely on Citrix to help them build simpler and more
cost-effective IT environments. Citrix partners with over 10,000 companies in more than 100 countries. Annual revenue in 2010 was $1.87 billion.

©2013 Citrix Systems, Inc. All rights reserved. Citrix, XenApp, XenDesktop, XenServer, NetScaler, NetScaler Gateway, FlexCast, HDX and Citrix
Receiver are trademarks or registered trademarks of Citrix Systems, Inc. and/or one or more of its subsidiaries, and may be registered in the United
States Patent and Trademark Office and in other countries. All other trademarks and registered trademarks are property of their respective owners.

1213/PDF citrix.com p